The problem with “AI” in business software
Most business owners have now seen an ERP add a chatbot. It answers general questions, sometimes invents numbers, and does little of the work that actually takes time. Meanwhile the real effort is still manual:
- Reading a drawing and typing out a BOM or BOQ.
- Copying a supplier’s invoice into a goods receipt and checking it against the PO.
- Working out what to order, and from whom.
- Chasing overdue invoices one by one.
- Pulling together figures for the GST return or the morning review.
The opposite risk is just as real: AI that posts records or sends messages on its own, so nobody knows why a number changed.
How AI fits into an ERP
AI is useful in an ERP in three places:
- Reading. Turning unstructured inputs — drawings, photos, PDFs, price lists, invoices, scope text — into structured lines.
- Drafting. Preparing a record a person would otherwise build by hand: a BOM, a BOQ, a purchase order, a GRN or the next RA bill.
- Answering. Explaining your own data in plain language, with tables and charts.
In each case the AI should work from your records, stay inside the user’s permissions, and hand a draft to a person. We set out this approach in AI agents that draft, people who decide.
What to look for in AI ERP software
- Drafts, not decisions. Nothing created, sent or posted without a person confirming.
- Permission-aware. The AI sees only what the user can see.
- Numbers from the ERP. Figures should come from your records, with the AI writing the explanation.
- Matching with review states. When AI reads a document, every line should be marked matched, needs review or no match.
- Metered and logged, so cost stays predictable and every call can be traced.
- Built into the workflow, not a separate chat window.
How to test AI claims in a demo
A short demo can make any AI look good. Bring your own material instead:
- A real drawing or scope document from a recent enquiry or tender. Compare the draft with what your estimator produced, line by line.
- A real supplier invoice with a short delivery or a price difference. Does the system flag it, or quietly accept it?
- A question about your own data — “which invoices are overdue by more than 60 days?” — asked by a user whose role should not see some of the answer.
- An approval path. Ask the vendor to show exactly where a person confirms the draft, and what happens if nobody does.
If the answers to these depend on the vendor’s own sample data, treat the AI as unproven for your business.
How Zaptiz uses AI
These principles apply across Zaptiz AI.
In Zaptiz Electrical
Zaptiz Electrical uses AI across estimating, purchase and finance:
- Drawing analysis of PDF and image drawings, with each line matched, flagged for review or left unmatched against your catalogue.
- Photo analysis of an existing panel, priced across low, mid and premium brand tiers.
- Quote Studio and the Quotation Drafting agent, turning an enquiry into a costed draft BOM and quotation.
- Price-List Intelligence, importing supplier price lists into the catalogue after review.
- Invoice & Delivery Reader, matching supplier documents to the PO and drafting the GRN for a three-way match.
- Smart Procurement, proposing purchases for shortages from the cheapest vendor.
- CRM Follow-up and Collections, drafting follow-ups and overdue reminders with a Pay-Now link.
- Production Planner, ranking open jobs and flagging late or material-blocked ones.
- GST Agent, computing and checking GSTR-1/3B figures — it never files.
- Daily Briefing and Ask-Your-Data, with answers and charts from your data, plus voice in the assistant.
In Zaptiz Construction
Zaptiz Construction applies the same model to projects:
- BOQ Generator, turning scope text or a messy Excel or PDF into an unpriced draft BOQ.
- Rate Suggestion, filling blank rate analysis coefficients from your library first.
- Procurement Agent, turning approved-but-unordered material into draft POs per vendor.
- Invoice & Challan Reader, matching a photo or PDF to the PO and drafting the GRN with differences flagged.
- RA Bill Drafting, preparing the next bill’s quantities with the over-billing check.
- Cost Overrun Watch, Labour & Plant Productivity, Collections and GST & TDS reports.
- Daily Site Briefing, a per-role morning brief. Construction briefings drop any sentence quoting a number they were not given.
Further construction agents — contract analysis, DPR summaries, quality and safety watch, cash-flow forecast and delay risk — are coming soon.
Controls
Agents draft; people approve. The assistant answers as the signed-in user, limited by role and project access. AI usage is metered in credits with per-user rate limits, and every call is logged. See Security and Using AI agents safely in Zaptiz.
Get started
Panel manufacturers can try every AI tool and agent in the 9-day free trial of Zaptiz Electrical, which includes 10,000 AI credits. Contractors can book a demo of Zaptiz Construction. Plans and credit packs are on the pricing page.